Back to all papers

Deep Learning-Based Denoising for High-Resolution Carotid Vessel Wall MRI Using Standard Neurovascular Coils.

December 19, 2025pubmed logopapers

Authors

Zeng L,Hsu YC,Wang L,Lu M,Keushkerian M,Nguyen KL,Johnson KJ,Altbach MI,Morris HD,DeMarco JK,Deshpande V,Mitsouras D,Saloner D,McNally JS,Kim SE,Roberts JA,Hadley JR,Parker DL,Treiman GS,Li D,Xie Y

Affiliations (11)

  • Biomedical Imaging Research Institute, Cedars-Sinai Medical Center, Los Angeles, California, USA.
  • Department of Bioengineering, UCLA, Los Angeles, California, USA.
  • Department of Cardiology, Radiology, and Bioengineering, UCLA, Los Angeles, California, USA.
  • VA Greater Los Angeles Healthcare System, Los Angeles, California, USA.
  • Department of Medical Imaging, University of Arizona, Tucson, Arizona, USA.
  • Department of Biomedical Engineering, University of Arizona, Tucson, Arizona, USA.
  • Walter Reed National Military Medical Center, Bethesda, Maryland, USA.
  • Siemens Medical Solutions, Austin, Texas, USA.
  • Department of Radiology and Biomedical Imaging, UCSF, San Francisco, California, USA.
  • Department of Radiology and Imaging Sciences, University of Utah, Salt Lake City, Utah, USA.
  • VA Salt Lake City, Salt Lake City, Utah, USA.

Abstract

To develop a deep learning (DL) denoising method to enhance high-resolution carotid vessel wall MRI quality acquired using a standard head-and-neck clinical coil. Fifty-five scans were performed as part of an ongoing multicenter study. Routine carotid VWI protocol including 2D T1- and T2-weighted TSE, 3D TOF-MRA, and MPRAGE was performed using simultaneous acquisition from a standard 20-channel head-and-neck coil and a high-sensitivity Neck-Shape-Specific (NSS) surface coil. Paired retrospective reconstructions with and without NSS coil elements served as the reference and input, respectively. A supervised DL model employing a residual UNet architecture was optimized and trained to map low-SNR inputs to high-SNR references, benchmarked against conventional denoising algorithms using quantitative and qualitative metrics. The DL denoiser substantially reduced noise while preserving vessel-wall structures across contrast-weighted sequences. It achieved PSNR > 31 dB and structural similarity index (SSIM) > 0.93 versus reference slices. In segmented vessel-wall and lumen regions of interest (ROIs), the DL approach achieved significantly higher SNR and CNR values than input images (p < 0.05), closely approaching the reference. Furthermore, inner-wall edge sharpness was maintained (Average ERD 7.50-8.51 mm with DL vs. 7.15-8.28 mm with references), supporting confident downstream plaques assessment. Radiologists' Likert ratings corroborated these image-quality improvements. A DL-based method was developed to improve high-resolution, multi-contrast carotid vessel wall MRI acquired using low-SNR standard head-and-neck coils. The resulting image quality was comparable to that obtained with specialized neck surface coils, potentially enabling broader access to advanced carotid imaging without the need for additional hardware.

Topics

Journal Article

Ready to Sharpen Your Edge?

Subscribe to join 7,400+ peers who rely on RadAI Slice. Get the essential weekly briefing that empowers you to navigate the future of radiology.

We respect your privacy. Unsubscribe at any time.